Financial institutions generate regulatory compliance reports such as Suspicious Activity Reports (SARs), Currency Transaction Reports (CTRs), and sanctions filings under strict legal requirements. These workflows remain highly manual, requiring analysts to review transaction data, draft narratives, and complete structured reports, resulting in significant cost and inefficiency. On average, a single report requires approximately two hours and costs around $75. This work presents ComplAI, a multi-agent system that automates key stages of the reporting pipeline while maintaining human oversight. The system consists of five specialized agents: Router, Aggregator, Narrative Generator, Evaluator, and PDF Filer and is supported by a hybrid knowledge architecture integrating Weaviate (semantic retrieval), PostgreSQL (deterministic validation and audit logging), and Redis (low-latency caching). Narrative generation leverages retrieval-augmented generation (RAG) and few-shot prompting, while validation combines rule-based checks with LLM-based scoring. ComplAI reduces report generation time to ~25 seconds (plus ~4 minutes review), lowers cost to ~$3 per report, and increases throughput to ~14 reports per hour (~27× improvement). The system augments, rather than replaces, human analysts, enabling scalable and auditable compliance automation. Duke University Durham, North Carolina April 2026
Mentor: Hengzhong Liu
Project poster (PDF)
